Heart sound segmentation using PCG signals

Abstract

Cardiac auscultation is a method used to listen heart sound. Condition of the heart can be predicted with cardiac auscultation because heart generates a specific rhythm of sound and any changes in the rhythm of the heart sound may be due to abnormalities of heart. Auscultation is an easy way to diagnose heart abnormalities; however, it needs training and years of physician’s experience to diagnose heart and identify any heart abnormalities. With years of experience it is still difficult to analyse heart sound. The ability to automatically identify abnormalities or at least support physician decision is relevant to ease the reach of medical diagnosis using mobile or Digi-scope. The phonocardiogram PCG signal are collected with the help of mobile or electronic stethoscope. Heart beat detection is very important in these signals for segmentation of fundamental heart sound. Finding heart rhythm in PCG signals is a challenging task due to the presence of noise i.e. external environmental noise or internal body noise. Another challenging task is segmentation of S1 and S2 heart sound. This thesis presents a novel approach for segmentation of S1 and S2 heart sounds by using some of heart sounds temporal and spectral features. Total of four features are extracted from these signals, in which two features are temporal feature and two are its spectral feature. K-mean clustering algorithm is used for segmentation of S1 and S2 on the bases of these features. PASCAL PCG heart sound dataset is used for testing our algorithm. Our method differentiates between S1 and S2 heart sounds to great extent and also improves the results.